{"id":"https://openalex.org/W4415797232","doi":"https://doi.org/10.52202/079017-1961","title":"Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels","display_name":"Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4415797232","doi":"https://doi.org/10.52202/079017-1961"},"language":"en","primary_location":{"id":"doi:10.52202/079017-1961","doi":"https://doi.org/10.52202/079017-1961","is_oa":false,"landing_page_url":"https://doi.org/10.52202/079017-1961","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 37","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2404.14313","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5046536754","display_name":"Jan-Philipp Fr\u00e4nken","orcid":"https://orcid.org/0000-0001-5467-1887","observed_orcids":["https://orcid.org/0000-0001-5467-1887"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jan-Philipp Fr\u00e4nken","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074923579","display_name":"Eric Zelikman","orcid":"https://orcid.org/0009-0001-7368-0460","observed_orcids":["https://orcid.org/0009-0001-7368-0460"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Eric Zelikman","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028522433","display_name":"Rafael Rafailov","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rafael Rafailov","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085625870","display_name":"Kanishk Gandhi","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kanishk Gandhi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039540205","display_name":"Tobias Gerstenberg","orcid":"https://orcid.org/0000-0002-9162-0779","observed_orcids":["https://orcid.org/0000-0002-9162-0779"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tobias Gerstenberg","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5001961716","display_name":"Noah D. Goodman","orcid":"https://orcid.org/0000-0002-9176-8802","observed_orcids":["https://orcid.org/0000-0002-9176-8802"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Noah Goodman","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.4436,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":{"value":0.85311222,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":90,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"61328","last_page":"61371"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.6859999895095825,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.6859999895095825,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.10360000282526016,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12072","display_name":"Machine Learning and Algorithms","score":0.07320000231266022,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/self-supervised-learning","display_name":"self-supervised learning","score":0.9340000152587891},{"id":"https://openalex.org/keywords/mutual-information","display_name":"mutual information","score":0.7369999885559082},{"id":"https://openalex.org/keywords/alignment","display_name":"alignment","score":0.382999986410141}],"concepts":[{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.5037000179290771},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4749000072479248},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4487000107765198},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.336899995803833},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.32739999890327454},{"id":"https://openalex.org/C181204326","wikidata":"https://www.wikidata.org/wiki/Q7239820","display_name":"Preference learning","level":3,"score":0.31130000948905945},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.28380000591278076},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.2621000111103058},{"id":"https://openalex.org/C26760741","wikidata":"https://www.wikidata.org/wiki/Q160402","display_name":"Perception","level":2,"score":0.23989999294281006},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2379000037908554}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/079017-1961","doi":"https://doi.org/10.52202/079017-1961","is_oa":false,"landing_page_url":"https://doi.org/10.52202/079017-1961","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 37","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2404.14313","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.14313","pdf_url":"https://arxiv.org/pdf/2404.14313","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2404.14313","doi":"https://doi.org/10.48550/arxiv.2404.14313","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2404.14313","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2404.14313","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.14313","pdf_url":"https://arxiv.org/pdf/2404.14313","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415797232.pdf","grobid_xml":"https://content.openalex.org/works/W4415797232.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7133218729","https://openalex.org/W4416037048","https://openalex.org/W4386755626","https://openalex.org/W4385572634","https://openalex.org/W7166887977","https://openalex.org/W4403852050","https://openalex.org/W4399399598","https://openalex.org/W6910572010","https://openalex.org/W4402671558","https://openalex.org/W4415316359"],"abstract_inverted_index":{"When":[0],"prompting":[1],"a":[2,14,37,40,62,86,93,128,145,154,160,213],"language":[3,64],"model":[4,10,41,65,129,148,157],"(LM),":[5],"users":[6],"often":[7],"expect":[8],"the":[9,74,97,132,206],"to":[11,13,72,172,183,195,205,218],"adhere":[12],"set":[15],"of":[16,193],"behavioral":[17],"principles":[18,35,175,203],"across":[19],"diverse":[20,173],"tasks,":[21],"such":[22,34],"as":[23],"producing":[24],"insightful":[25],"content":[26],"while":[27],"avoiding":[28],"harmful":[29],"or":[30,52,70,226],"biased":[31],"language.":[32],"Instilling":[33],"(i.e.,":[36],"constitution)":[38],"into":[39],"is":[42],"resource-intensive,":[43],"technically":[44],"challenging,":[45],"and":[46,80,91,106,121,181,199],"generally":[47],"requires":[48,127],"human":[49,227],"preference":[50,68,223],"labels":[51,69],"examples.":[53],"We":[54],"introduce":[55],"SAMI,":[56],"an":[57,112],"iterative":[58],"algorithm":[59],"that":[60,130,188,212],"finetunes":[61],"pretrained":[63,99,147,214],"(without":[66],"requiring":[67],"demonstrations)":[71],"increase":[73],"conditional":[75],"mutual":[76],"information":[77],"between":[78,104,119],"constitutions":[79,151,220],"self-generated":[81],"responses":[82],"given":[83],"queries":[84],"from":[85],"dataset.":[87],"On":[88],"single-turn":[89,124],"dialogue":[90],"summarization,":[92],"SAMI-trained":[94],"mistral-7b":[95],"outperforms":[96],"initial":[98],"model,":[100],"with":[101,116],"win":[102,117,162,191],"rates":[103,118,192],"66%":[105],"77%.":[107],"Strikingly,":[108],"it":[109,189],"also":[110],"surpasses":[111],"instruction-finetuned":[113,156],"baseline":[114],"(mistral-7b-instruct)":[115],"55%":[120],"57%":[122],"on":[123,137,164],"dialogue.":[125],"SAMI":[126,170],"writes":[131],"principles.":[133],"To":[134],"avoid":[135],"dependence":[136],"strong":[138,146],"models":[139,185],"for":[140,197,201],"writing":[141],"principles,":[142],"we":[143,167],"align":[144],"(mixtral-8x7b)":[149],"using":[150,222],"written":[152],"by":[153],"weak":[155],"(mistral-7b-instruct),":[158],"achieving":[159],"65%":[161],"rate":[163],"summarization.":[165],"Finally,":[166],"investigate":[168],"whether":[169],"generalizes":[171],"summarization":[174],"(e.g.,":[176],"\"summaries":[177],"should":[178],"be":[179],"scientific\")":[180],"scales":[182],"stronger":[184],"(llama3-70b),":[186],"finding":[187],"achieves":[190],"up":[194],"68%":[196],"learned":[198],"67%":[200],"held-out":[202],"compared":[204],"base":[207],"model.":[208],"Our":[209],"results":[210],"show":[211],"LM":[215],"can":[216],"learn":[217],"follow":[219],"without":[221],"labels,":[224],"demonstrations,":[225],"oversight.":[228]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":1}],"updated_date":"2026-10-12T09:40:04.860316","created_date":"2025-11-03T00:00:00"}